* refactor: move tests for context function added in apollo-server into integration test suite
* Move existing context tests in integration testsuite into `describe('context field'`
# Conflicts:
# packages/apollo-server-integration-testsuite/src/ApolloServer.ts
* Add back tests in apollo-server along with comments to explain the duplication
This reverts commit 25782382b8.
Unfortunately, `npm publish` isn't okay with using a `README.md` from
outside the package's root, even if the README at the root of the repository
is far more reasonable to use for the `apollo-server` package itself.
While it used to be required to pass the `--save` flag in order to actually
write the dependency to the project's `package.json` file, this is no longer
the case in recent versions of npm.
Previously, in order to only load dependencies which only work on Node.js
and fail in non-Node.js environments, we introduced a check which did a
string-comparison of `process.release.name` to see if it is was `node`.
This was first introduced in 9c563fae50 as
part of #2054, but has gone on to be useful for other purposes as well.
Some Node.js forks such as NodeSource's N|Solid, which is a fork of Node.js
which follows-up each Node.js release with a custom build that includes
additional native addons but is otherwise the same, override this value with
their own name. This means that N|Source returns `nsolid`, despite the fact
that it is almost entirely the same as Node.js.
Luckily, N|Solid leaves the base version of its Node.js in
`process.versions.node` (and additionally adds its own
`process.versions.nsolid`). By relaxing the string comparison on
`process.release.name`, we should still be able to accurately detect the
environment we want - which is "Close enough to Node.js!".
Fixes https://github.com/apollographql/apollo-server/issues/2356
* Add AS constructor with context usage to tests
* fix(typing): Context template type should be 'object' and not 'any'
```context?: ContextFunction<ExpressContext, any> | Context<any>;```
is
```context?: ContextFunction<ExpressContext, any> | any;```
which is
```context?: any;```
https://github.com/Microsoft/TypeScript/issues/18568
Since the ContextFunction<...> part is ignored, no type hinting will be provided when a function is passed to `context`. We don't really mean `any`, we mean an object with any shape, so we should use `object` instead
* fix(typing): type ContextFunction params separately from resulting context object
* Update changelog
* rename: ContextContent -> ProducedContext
* Type formatErrors function to specify that it accepts and returns ApolloError
* Update changelog
* type(formatErrors): accept GraphQLError, return GraphQLFormattedError
* update changelog
As we're moving the community from Slack to Spectrum, its time to update
these shields from our project README files (which show on npm and in
GitHub) to reflect the new home for conversation.
If you haven't already joined the new Apollo community on Spectrum.chat,
visit https://spectrum.chat/apollo/ to join the discussion!
For more details on the move, checkout the Apollo blog post by @hwillson:
https://blog.apollographql.com/goodbye-slack-hello-spectrum-8fa6b979645b
This is necessary after the changes in 0f75909e35
and https://github.com/apollographql/apollo-server/pull/2344, as the
`AddressInfo` type didn't become available until a future version of
`@types/node`.
Regardless, it seems to me that this should be inferred without any problem
and we don't actually rely on the `AddressInfo` type, and instead are only
looking for an interface which has `port`, `address` and `family`, as the
inferred return value from `http.Server.prototype.address()` does.
While it would be perfectly fine to use `import *` in this particular case,
we've moved away from the star-import pattern in this repository, instead
preferring the `esModuleInterop` compiler option which seems to be the
direction that TypeScript is moving as it's now the default with `tsc --init`.
TypeScript clearly lost its battle with `import *` which many consider to
have been an incorrect direction in the first place. Some build tools won't
work with star imports, and using them can make some optimizations more
difficult. By avoiding this pattern here, hopefully we'll avoid moving back
in that direction inadvertently.
Here's a read I found at one point which roughly explains:
https://itnext.io/great-import-schism-typescript-confusion-around-imports-explained-d512fc6769c2
We've generally moved away from the star-import pattern which TypeScript
also no longer generates with its default `tsc --init` configuration, so it
seems to make sense to keep picking exactly what we want since this pattern
is no longer forced upon us by TypeScript.
* typing: context function could return value synchronously
* Create ApolloServerExpressConfig type that has req and res in context
* Have apollo-server-express's constructor also use the express-specific config
* Update changelog
These missing configurations, which weren't identified in the original
implementation of `apollo-server-azure-functions` in #1926, are responsible
for the failures which have surfaced in the #2228, which updates Jest to v24.x.
* test: ensure "total duration" > "duration of resolvers"
Add test for https://github.com/apollographql/apollo-server/pull/2298
* Update tracing duration test to be more comprehensive
Previously we were only guaranteeing that individual resolvers didn't have a duration longer than the total duration. The updated test guarantees that the duration (from when the first resolver started to when the last resolver ended) should be less than total duration
* Add another resolver to tracing test
* Prefer `const` over `let` for variables which aren't mutated.
The apollo-server-core package uses Node's built-in crypto module only to
create SHA-256 and -512 hashes.
When we're actually running in Node, the native crypto library is clearly
the best way to create these hashes, not least because we can assume it
will be available without having to bundle it first.
Outside of Node (such as in React Native apps), bundlers tend to fall back
on the crypto-browserify polyfill, which comprises more than a hundred
separate modules. Importing this polyfill at runtime (likely during
application startup) takes precious time and memory, even though almost
all of it is unused.
Since we only need to create SHA hashes, we can import the much smaller
sha.js library in non-Node environments, which happens to be what
crypto-browserify uses for SHA hashing, and is a widely used npm package
in its own right: https://www.npmjs.com/package/sha.js.
When getEngineServiceId does not return a truthy value, we can save memory
and startup time by avoiding importing apollo-engine-reporting, which
(among its other dependencies) imports the protobuf.js implementation.
